Diagnostic, not prescriptive: the four things a reading may move.

The grant

A closed list is a stronger instrument than a vague one.

Most systems describe their analytical components loosely — the diagnostics inform the process, the readings support decision-making — and loose descriptions license anything. MARS names four permitted effects instead. Confidence: how firmly the operator holds their existing plan. Review priority: what gets examined first and most carefully at the weekly and monthly cadence. Throttle discussion: whether the conversation about capacity and pace is even opened this cycle. Risk posture: the general defensive or constructive stance held inside whatever range the gate has already fixed. Four verbs, each attached to something the operator does rather than something the account does. The list's value is in what it excludes by being finite.

Why deployment is excluded

Deployment answers a different question with a different cost of error.

The exclusion is not a hedge against the engine being unreliable. It follows from the two layers answering genuinely different questions. Deployment authority asks what the account can survive, and it takes drawdown as its input because survival is a capital-state question. The SDE asks whether the machine's behaviour is improving or deteriorating, and it takes performance structure as its input because quality is a behaviour question. These come apart routinely — a system can be structurally improving while deep in drawdown, and structurally deteriorating at an equity high — and when they do, the survival question has to be the one that binds, because being wrong about quality costs a suboptimal month while being wrong about survival can cost the account. Letting a quality reading move a survival decision would double-count the same evidence and resolve it toward the more optimistic of two conclusions.

• Gate input is drawdown. SDE input is structural performance. Different evidence, different question.
• The asymmetry of consequences is what decides which layer binds when they disagree.
• Excluding deployment is what allows the SDE to publish encouraging readings without creating pressure.

Throttle discussion, precisely

The third item is a conversation, not a setting.

Of the four permitted effects, throttle discussion is the one most easily misread, because the Throttle Control Panel does hold real operational power and the wording seems to hand the SDE a lever on it. It does not. The grant is over whether the discussion is opened — whether the structural picture justifies putting capacity on the agenda this cycle — and not over what the discussion concludes. The throttle takes its own inputs, applies its own logic, and reaches its own selections; a structural diagnosis can put the question on the table and has no vote once it is there. The distinction is the same one that runs through the whole rung: the engine changes what gets examined, and other components decide what happens as a result.

In practice

What the rule looks like on a Saturday.

A structural reading arrives showing RAER degrading on the Trend No-Partial branch with high conviction and stable drift. Under the closed list, four things may follow. Confidence in the current allocation to that branch drops. The branch moves to the top of the review order, ahead of metric families that look fine. The throttle question is legitimately opened, since sustained efficiency degradation on a high-variance branch is exactly the sort of structural fact capacity discussion exists for. And posture within the existing tier tilts defensive. What may not follow is a size cut applied on the spot, a self-imposed branch suspension, or a decision to trade the next tier down because the reading felt serious. Those are deployment actions, they belong to the gate and the plan, and reaching for them is how a diagnostic layer quietly becomes an unaccountable one.

The key idea

Naming what a component may do is how you find out what it did.

A closed list makes overreach visible. When the permitted effects are four named things, an action outside them stands out immediately and can be traced to the reading that produced it, which means the boundary can actually be audited rather than merely asserted. Vague grants of influence cannot be violated, which sounds like freedom and is really just an absence of accountability.